Paper Title
Short Term Voice Traffic Forecast in 3G/UMTS Networks Using Machine Learning and Statistical Methods

Abstract
This study aims to develop models for precisely forecasting the voice traffic of a commercially deployed Third Generation Universal Mobile Telecommunication System (3G/UMTS) network in Turkey using various machine learning and statistical methods. Particularly, three machine learning methods including Support Vector Machine (SVM), Multilayer Perceptron (MLP) and Random Forest (RF), and a statistical method called Holt Winters (HW) have been utilized to build the forecasting models. The dataset used in this study comes from a real network, and contains 2112 hourly basis voice traffic captured between October 2015 and January 2016. The performance of the forecasting models has been evaluated by computing the mean absolute percentage error (MAPE), a popular metric commonly used in forecasting applications. The results show that SVM yields the lowest MAPE�s ranging from 7.82% to 22.18% and can be utilized for forecasting the 3G/UMTS voice traffic. Keywords� Machine Learning, Voice Traffic Forecast, 3G/UMTS, Time Series Forecasting.